Classifying input examples using a comparison set
Methods, systems, and apparatus for classifying a new example using a comparison set of comparison examples. One method includes maintaining a comparison set, the comparison set including comparison examples and a respective label vector for each of the comparison examples, each label vector including a respective score for each label in a predetermined set of labels; receiving a new example; determining a respective attention weight for each comparison example by applying a neural network attention mechanism to the new example and to the comparison examples; and generating a respective label score for each label in the predetermined set of labels from, for each of the comparison examples, the respective attention weight for the comparison example and the respective label vector for the comparison example, in which the respective label score for each of the labels represents a likelihood that the label is a correct label for the new example.
1. One or more non-transitory computer storage medium encoded with instructions that, when executed by one or more computers, cause the one or more computers to perform operations for classifying a new image using a comparison set of comparison images, the operations comprising:
maintaining a comparison set, the comparison set comprising a plurality of comparison images and a respective label for each of the plurality of comparison images;
receiving a new image; and
determining a new label for the new image based on (i) the new image and the plurality of comparison images and (ii) the labels of the plurality of comparison images, comprising applying an attention mechanism to the new image and the plurality of comparison images.
2. A method performed by one or more computers for classifying a new image using a comparison set of comparison images, the method comprising:
maintaining a comparison set, the comparison set comprising a plurality of comparison images and a respective label for each of the plurality of comparison images;
receiving a new image; and
determining a new label for the new image based on (i) the new image and the plurality of comparison images and (ii) the labels of the plurality of comparison images, comprising applying an attention mechanism to the new image and the plurality of comparison images.
3. The method of claim 2 , wherein determining the new label for the new image based on (i) the new image and the plurality of comparison images and (ii) the labels of the plurality of comparison images comprises:
determining, for each comparison image in the plurality of comparison images, a respective attention weight by applying a neural network attention mechanism to the new image and to the comparison image; and
determining the new label for the new image based on the respective attention weights determined for the plurality of comparison images and the labels of the plurality of comparison images.
4. The method of claim 3 , wherein the new label comprises a respective label score for each label of a plurality of labels, the respective label score representing a likelihood that the label is a correct label for the new image.
5. The method of claim 3 , wherein determining the respective attention weight for each comparison image comprises:
processing the comparison image using a comparison image neural network to determine a numeric embedding of the comparison image; and
processing the new image using a new image neural network to determine a numeric embedding of the new image.
6. The method of claim 5 , wherein the comparison image neural network is the same as the new image neural network.
7. The method of claim 5 , wherein determining the respective attention weight for each comparison image further comprises:
determining the respective attention weight by determining a similarity metric between the numeric embedding of the comparison image and the numeric embedding of the new image.
8. The method of claim 7 , wherein the similarity metric is a cosine distance.
9. The method of claim 5 , wherein the comparison image neural network is a bidirectional Long-Short Term Memory (LSTM) neural network configured to process the comparison image in a context of a sequence of comparison images from the comparison images in the comparison set to generate the numeric embedding of the comparison image.
10. The method of claim 5 , wherein the new image neural network is an LSTM neural network with read attention over the numeric embeddings for the plurality of comparison images that is configured to process a sequence comprising K instances of the new image to generate the numeric embedding of the comparison image.
11. The method of claim 5 , wherein the comparison image neural network is a convolutional neural network configured to receive an input comparison image and to generate the embedding of the comparison image.
12. The method of claim 5 , wherein the new image neural network is a convolutional neural network configured to receive an input new image and to generate the embedding of the new image.
13. The method of claim 5 , wherein the new image neural network and the comparison image neural network share at least some parameters.
14. The method of claim 3 , wherein the neural network attention mechanism has been learned by training on a training set of images different from the comparison set.
15. A system comprising one or more computers and one or more non-transitory computer storage medium encoded with instructions that, when executed by the one or more computers, cause the one or more computers to perform operations for classifying a new image using a comparison set of comparison images, the operations comprising:
maintaining a comparison set, the comparison set comprising a plurality of comparison images and a respective label for each of the plurality of comparison images;
receiving a new image; and
determining a new label for the new image based on (i) the new image and the plurality of comparison images and (ii) the labels of the plurality of comparison images, comprising applying an attention mechanism to the new image and the plurality of comparison images.
16. The system of claim 15 , wherein the operations for determining the new label for the new image based on (i) the new image and the plurality of comparison images and (ii) the labels of the plurality of comparison images comprise:
determining, for each comparison image in the plurality of comparison images, a respective attention weight by applying a neural network attention mechanism to the new image and to the comparison image; and
determining the new label for the new image based on the respective attention weights determined for the plurality of comparison images and the labels of the plurality of comparison images.
17. The system of claim 16 , wherein the operations for determining the respective attention weight for each comparison image comprise:
processing the comparison image using a comparison image neural network to determine a numeric embedding of the comparison image; and
processing the new image using a new image neural network to determine a numeric embedding of the new image.
18. The system of claim 17 , wherein the new image neural network and the comparison image neural network share at least some parameters.
19. The system of claim 17 , wherein the operations for determining the respective attention weight for each comparison image further comprise:
determining the respective attention weight by determining a similarity metric between the numeric embedding of the comparison image and the numeric embedding of the new image.
20. The system of claim 19 , wherein the similarity metric is a cosine distance.